Codex desktop app (Store 26.924.2738.0) exits silently ~2s after launch on Windows 11 build 26200

The desktop app launches and then kills itself about 2 seconds later. No error dialog, no crash, no log line. From the user’s side it just looks like double-clicking does nothing.

This has been reproduced dozens of times, including after a reboot and after a clean reinstall.

Environment

  • App: ChatGPT / Codex desktop app, Microsoft Store (MSIX), package OpenAI.Codex
  • PackageFullName: OpenAI.Codex_26.924.2738.0_x64__2p2nqsd0c76g0
  • Version: 26.924.2738.0
  • OS: Windows 11 Pro, 10.0.26200 (Build 26200), x64
  • Manifest declares MinVersion 10.0.19041.0 / MaxVersionTested 10.0.26100.0
    → my OS build (26200) is newer than the declared MaxVersionTested (26100)
  • Bundled runtime: codex-cli 0.158.0-alpha.2.1
  • The companion service CodexSandboxService.OpenAI.Codex is installed and running

What the logs show

Activation succeeds every single time — TWinUI/Operational event 1621:

Attempted activation of the app, 0, operation completed successfully.

Then the AppX container is built and torn down ~2s later — AppModel-Runtime/Admin:

event 210  created desktop AppX container
event 201  created process for OpenAI.Codex_2p2nqsd0c76g0!App
event 217  destroyed desktop AppX container      <- ~2 seconds later

Process sampling: 13 ChatGPT.exe processes exist at t+1.5s, 0 processes at t+5.5s.

The one error dialog, seen only on the very first launch, never again

Unable to load organization settings
The app is paused, waiting for secure organization settings to load.
Network connectivity, proxy, or firewall may be preventing access.
Check your network connection and try again, or sign out and use a different account.
If the problem persists, contact your IT administrator.
[Retry] [Sign out] [Exit]

That looks like a fail-closed startup gate: if the managed-configuration / organization-settings fetch can’t complete, the app refuses to run. Once it started exiting silently, the dialog never came back.

No usable error output anywhere

  • %LOCALAPPDATA%\CrashDumps: no matching dump
  • Windows event log Application: no Application Error / WER entry
  • Chromium Crashpad folder: no .dmp
  • ~/.codex/logs_*.sqlite: the logs table is empty
  • %APPDATA%\Codex and %LOCALAPPDATA%\Codex exist but stay empty

Already ruled out — all tested, all negative

  • Network: with a working proxy, api.openai.com/v1/models returns 401 and chatgpt.com/backend-api/me returns 403. Still exits.
  • Model channel: codex.exe exec --skip-git-repo-check "reply with exactly: OK" returned a real answer in 14.5s (8,221 tokens). Engine → config.toml → local relay → upstream works end to end.
  • codex doctor: 19 ok · 1 idle · 5 notes · 5 warn · 0 fail (degraded). Its only concrete warnings are a reachability note and an HTTP 502 on its own custom-API route probe — contradicted by the successful real call above.
  • CODEX_HOME env var (was pointing at a deleted folder; removing it changed nothing)
  • ~/.codex contents, config.toml content (replaced with a minimal config), Chromium profile (renamed away)
  • Full uninstall + reinstall from the Store (same version, still exits) and Settings → Apps → Reset
  • Rebooting Windows; disabling Memory integrity / HVCI
  • Third-party AV: only Windows Defender is present
  • Exploit Protection / IFEO hijacking: system mitigations all NOTSET, no IFEO entry for ChatGPT.exe
  • Smart App Control: off
  • Restarting explorer.exe
  • Four different launch paths: desktop shortcut (AUMID), shell:AppsFolder\..., running app\ChatGPT.exe directly, and codex app <path> — all identical.

Questions

  1. Is version 26.924.2738.0 known to be incompatible with Windows 11 build 26200?
  2. Is there any way to bypass or soften that startup organization / managed-configuration check? Right now it seems to be fail-closed, and there is no way to get into the app at all.
  3. Is anyone else on build 26200 seeing the same thing?

Thanks — happy to provide any further logs you want.

Exact same … to the letter